Extending Appearance Based Gait Recognition with Depth Data
Round 1
Reviewer 1 Report
This paper proposed two combinations of parameters that can be performed for the improvement of gait recognition with depth data.
Lines 32-33. “To capture the depth information structured light or time-of-flight technologies are commonly used. In both cases an active low power IR projector….” Unclear. Please consider revising. What are the both cases? Also, a comma can be used after the word cases: In both cases, an active low…. Line 100. What is Kernel SVM? Is SVM= Support vector machines? If yes please mention that SVM is a notation. The actual algorithm that process the data and the considered parameters is kNN (as stated in line 336). However, no discussion of this aspect is presented in the manuscript. Please motivate why this algorithm was used. Overall, there are some small language issues. Also, adding some commas to separate some sub-sentences would improve the readability of the manuscript.Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The core step of the paper consist of two extensions of gait recognition combining data from RGB video images and depth data from 3D sensors. The extensions are simple, but they seem robust and interesting from a practical standpoint. Results seem convincing. Literal presentation is good, but there is room for improvement in some parts of the paper text as well as figures and tables. In addition, experimental section might be improved, extending the number of experiments and classification methods. In summary, I consider the contents of the paper are potentially publishable, but the following issues should be addressed in a revised version of the paper.
- Although, in general, the paper is well written, there are some typos and writing style issues in the paper, e.g., (i) Abstract, “appearance-based recognition methods, this” instead of “appearance based recognition methods this”; (ii) page 2, please include a brief definition of the two proposed methods. “In this work two approaches are proposed….” (iii) page 3, please state a rough classification of the methods reviewed and joint some of the paragraphs; (iv) page 4, “Moreover, identical sensors” instead of “Moreover identical sensors,”; (v) definitions of “N” and “t” could be eliminated since they were defined before; (vi) page 7, Section 4, change ““two methods are proposed“ to “two methods are proposed (Height-GEI integration, HGEI-I and Height-GEI integration, HGEI-f)”; (vii) page 7, Section 4, “the initial processing steps”, please recall here those steps and sections where are defined; (viii) page 10, caption of Figure 8, change “Example RGB” to “Example of RGB”; (ix) page 12, check this “presented by in”; (x) page 12, Section 5.1.5, explain how the weights were varied?.
Therefore, a complete English proofreading of the paper is recommended.
- There are several classification and fusion methods that can easy implemented. I suggest the authors to include other classification and/or fusion methods to provide competitive results for comparison with the ones of k-nearest neighbors (kNN). In addition, a rationale about the selection of kNN is needed.
- In order to evaluate comprehensively the performance of the method, more experiments should be added, for instance, by randomly changing the partition training/testing of the records. This would allow the stability and variance of the obtained results to be measured. Besides, please consider to implement other performance indexes such as balanced accuracy and kappa.
- The results of Tables 2, 3, and 4 are difficult to read. These tables could be summarized in some figures, or leave some table and the rest of the content in figures or text, depending on whether the results are similar.
- The sum rule information fusion method was applied to fuse the results from the single methods. The weights were estimated in a rather experimental manner obtaining a suboptimal solution. Recently, alpha integration has been proposed for optimal soft fusion of several classifiers. Alpha integration provides weights for an optimum linear combination with respect to criteria such as least mean squares (LMSE) or the minimum probability of error (MPE). Please include more competitive fusion methods and/or discuss about competitive methods for the fusion stage of the procedure proposed in the paper. I suggest the following references: (i) Safont G., Salazar A., Vergara L. Multiclass Alpha Integration of Scores from Multiple Classifiers. Neural Computation, 2019, 31(4): 806-825; (ii) Mohandes M., Deriche M., Aliyu S.O. Classifiers combination techniques: a comprehensive review. IEEE Access, 2018, 6: 19626 – 19639. You can find definitions of “early fusion” and “late fusion” in reference (ii) that are related with the two proposed extensions (HGEI-I and HGEI-f, respectively). Please comment that in the paper.
- Besides the implementation of advanced methods of fusion, another line of future work would be the extension of the results to multimodal fusion considering audio, for instance, RGB, and depth recordings.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper deals with a method which extends the Gait Recognition using depth data.
The work is well written and the method is well conducted. Also, the interest for readers of the Journal is relevant on this topic. In my opinion authors should try to be more consistent in stating the the main objectives of the paper since the reader understand appropriately which are the main aims of the paper only at the beginning of SSection 4. In other words a partial rewriting of paragraph from line 55 to 57 is, in my opinion, required to be more effective.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The quality of the paper has been improved significantly. All my concerns have been adequately addressed including the following: improving of literal presentation; implementing competitive methods of classification; implementing an advanced fusion method; extending the number of experiments; and including more performance evaluation indexes. Therefore, I consider the contents of the paper should be ready for publication.
